Interest rate risk of Chinese commercial banks based on the GARCH-EVT model

Interest rate volatility significantly impacts asset pricing and risk management, prompting research into predicting interest rate market volatility. China's expanding market and rapid development of ESG assets and investments, stock markets, and capital markets have drawn global attention. The COVID-19 pandemic and subsequent economic downturn, coupled with China's 2020 lockdowns, increased economic risks and amplified interest rate fluctuations. This poses challenges to Chinese commercial banks, still operating under a bank-oriented financial system, increasing their interest rate risk. Existing research on commercial banks' interest rate fluctuation is limited, mainly focusing on interest rate liberalization's effects. However, market-determined interest rates increase risks for banks, highlighting the need for improved models to predict interest rate fluctuations and manage risks effectively. This study aims to address this gap by developing a more reliable model to minimize commercial banks' interest rate risks, contributing to stable operation and reduced systemic financial risk. The research questions are: (1) What are the risks commercial banks face due to interest rate fluctuations? (2) How can these risks be minimized using an interest rate fluctuation model? (3) What policies are needed to guide commercial banks in managing interest rate risks? The study's contributions include providing a foundation for future research, exposing stakeholders to market risks, and offering strategies for policymakers to minimize risks and maximize returns for commercial banks in China.

Value-at-Risk (VaR) is a common measure for interest rate risk. However, the historical simulation method, widely used, struggles to accurately predict risks during periods of high volatility. GARCH models, which capture the volatility clustering effect, have been employed to improve VaR estimation. However, assuming a normal distribution for returns may underestimate risks due to the infrequent but significant impact of extreme changes. To address this, researchers have incorporated more flexible distributions like Student's t, GED, and NIG into GARCH models. Various GARCH models (EGARCH, TGARCH, APARCH, FIGARCH, MNGARCH, BEKK, and DCC) have been developed to account for leverage effects and long-term dependence. Many studies have applied GARCH models to different markets, like precious metals and cryptocurrencies. The challenge remains in selecting the most appropriate GARCH model. Extreme Value Theory (EVT), particularly the Peaks Over Threshold (POT) method, provides a solution by directly modeling the tail distribution. EVT avoids the need to assume a specific distribution for the entire dataset, enabling more accurate risk estimation during extreme events. Studies have successfully combined GARCH and EVT to estimate VaR in various fields, but their application in the Chinese commercial banking sector's interest rate risk is lacking.

This study uses the Shanghai Bank Overnight Offered Rate (O/N SHIBOR) from October 24, 2015, to October 22, 2021. The first-order logarithmic difference of the O/N SHIBOR series is calculated to eliminate non-stationarity. The Ljung-Box (LB) test checks for autocorrelation in the series. A Markov regime switching (MS) model is applied to identify regime states (high and low volatility). Value-at-Risk (VaR) is calculated using both parametric methods (with various GARCH models and distributions: Normal, skewed normal, Student's t, skewed Student's t, GED, skewed GED, and NIG) and the EVT method. ARMA models are used to model the mean equation of the GARCH models. The following GARCH models are used: standard GARCH(1,1), EGARCH(1,1), and TGARCH(1,1). The EVT method utilizes the generalized Pareto distribution (GPD) to model the tail behavior of the data. VaR is calculated based on the GPD parameters. The Kupiec and Christoffersen tests backtest the VaR models. Finally, a robustness test is conducted by changing the sample period.

The Markov regime switching model reveals a clear shift in SHIBOR volatility around 2018, transitioning from a low-volatility to a high-volatility regime. The EGARCH-GED model exhibits the lowest AIC value among the GARCH models tested, suggesting it best fits the data. The EGARCH model's negative α parameter and the TGARCH model's significant γ parameter indicate an asymmetric effect, with volatility increasing more significantly after negative shocks. The parameter β values in both EGARCH and TGARCH models exceed 0.95, indicating significant volatility clustering. The residual analysis shows that the EGARCH-GED model's residuals are close to white noise. The EVT method, combined with the EGARCH-GED model, shows improved VaR estimation at the 99% confidence level (-0.6609%), whereas GARCH models are superior at lower confidence levels. The backtesting results using the Kupiec and Christoffersen tests demonstrate the validity of the GARCH-EVT model at the 99% and 95% confidence levels. Robustness tests using different sample periods confirm the EGARCH-GED model's suitability.

The study's findings address the research questions by demonstrating the increased interest rate risk faced by Chinese commercial banks due to market-determined interest rates. The combination of GARCH and EVT models proves superior in estimating VaR, especially at higher confidence levels, reflecting the importance of considering extreme events. The study highlights the asymmetric effect of positive and negative shocks on volatility. The volatility clustering effect is pronounced, implying that current fluctuations have a substantial impact on future volatility. The results are partially inconsistent with previous research using different models and time periods. The superior performance of the GARCH-EVT model provides a more comprehensive and accurate measure of interest rate risk compared to traditional GARCH models alone.

This paper successfully employed a GARCH-EVT model to analyze interest rate risk in Chinese commercial banks post-liberalization. The EGARCH-GED model emerged as the most suitable, accurately capturing volatility clustering and asymmetry. EVT improved VaR estimation at the 99% confidence level. The findings underscore the need for robust risk management strategies. Future research could explore more sophisticated models, incorporate macroeconomic factors, and consider CVaR for a more comprehensive risk assessment. Policy recommendations focus on enhancing risk awareness, establishing robust pricing mechanisms, creating dedicated risk management departments, and strengthening employee training.

The study's limitations include the potential for future economic and societal changes to affect the model's accuracy. The use of a single benchmark interest rate might not fully capture the complexity of interest rate risk. Further research using more advanced models, including multi-regime analysis and CVaR, could provide more robust results. The focus on O/N SHIBOR might not fully capture the risk profile of all interest rate-sensitive products held by commercial banks. Finally, the model’s performance might change with time, requiring frequent updates and recalibration.